Re: Strange results of DIV_ROUND_CLOSEST

[Guenter Roeck <linux@roeck-us.net>]

On Tue, Dec 18, 2012 at 10:04:56PM +0100, Juergen Beisert wrote:
> Hi Guenter,
> 
> Guenter Roeck wrote:
> > On Tue, Dec 18, 2012 at 04:03:41PM +0100, Juergen Beisert wrote:
> > > commit 263a523d18bca306016d75f5c8d5c57c37fe52fb changes the code of
> > > DIV_ROUND_CLOSEST in include/linux/kernel.h to fix a compile time
> > > warning.
> > >
> > > But now feeding in a zero into this macro results into 4198403. Tested
> > > with gcc 4.4.3 and 4.7.2, on arch x86 and ARM.
> > >
> > > I can reproduce this behaviour, when my ADC delivers a '0' value in the
> > > driver drivers/hwmon/s3c-hwmon.c in function s3c_hwmon_ch_show() with a
> > > current 3.7.1 kernel. The value is correct again, when the ADC delivers
> > > at least a '1'.
> > >
> > > Any ideas how to fix it correctly?
> >
> > Odd one. I ran the macro through a large number of values and divisors as
> > well as various optimization options, with different compilers, and always
> > get correct results.
> >
> > What are your compile options, and what are the channel multiplier and
> > dividers set to ?
> 
> Refer the lines 177 to 182 in drivers/hwmon/s3c-hwmon.c. "cfg->mult" is '3300' 
> in my case, and "cfg->div" is '1023'. And whenever s3c_hwmon_read_ch() 
> returns '0' line 184 returns '4198403' since Linux-3.6. checked with my 
> gcc-4.6.2 cross compiler for Linux-3.6 and with gcc-4.6.2 for Linux-3.7.
> 
> I did a quick test with this macro on my host with gcc-4.4.3 and a simple 
> userland program and surprise, surprise:
> 
> result = DIV_ROUND_CLOSEST(0, 1023);
> 
> works as expected (result is 0), but
> 
> int x = 0;
> unsigned y = 1023;
> result = DIV_ROUND_CLOSEST(x, y);
> 
> gives me result = 4198403!
> 

Juergen,

Can you test the following patch ?

Thanks,
Guenter

>From d4a639c6ace7cb0de247f59c38abac72d671f135 Mon Sep 17 00:00:00 2001
From: Guenter Roeck <linux@roeck-us.net>
Date: Tue, 18 Dec 2012 15:48:44 -0800
Subject: [PATCH] linux/kernel.h: Fix DIV_ROUND_CLOSEST for unsigned divisors

Commit 263a523 fixes a warning seen with W=1 due to change in
DIV_ROUND_CLOSEST. Unfortunately, the C compiler converts divide operations
with unsigned divisors to unsigned, even if the dividend is signed and
negative (-10 / 5U = 858993457). As a result, DIV_ROUND_CLOSEST(0, 2U)
and similar operations now return bad data.

Fix by checking for the divisor variable type when deciding which operation
to perform. This fixes DIV_ROUND_CLOSEST(0, 2U), but still returns bad data
for negative dividends divided by unsigned divisors. Mark the latter case as
unsupported.

Reported-by: Juergen Beisert <jbe@pengutronix.de>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
---
An alternative would be to always convert divisors to signed:
 	int __d = (int)divisor;			\
Not sure if that would be better.

 include/linux/kernel.h |    6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/include/linux/kernel.h b/include/linux/kernel.h
index d97ed58..45726dc 100644
--- a/include/linux/kernel.h
+++ b/include/linux/kernel.h
@@ -77,13 +77,15 @@
 
 /*
  * Divide positive or negative dividend by positive divisor and round
- * to closest integer. Result is undefined for negative divisors.
+ * to closest integer. Result is undefined for negative divisors and
+ * for negative dividends if the divisor variable type is unsigned.
  */
 #define DIV_ROUND_CLOSEST(x, divisor)(			\
 {							\
 	typeof(x) __x = x;				\
 	typeof(divisor) __d = divisor;			\
-	(((typeof(x))-1) > 0 || (__x) > 0) ?		\
+	(((typeof(x))-1) > 0 ||				\
+	 ((typeof(divisor))-1) > 0 || (__x) > 0) ?	\
 		(((__x) + ((__d) / 2)) / (__d)) :	\
 		(((__x) - ((__d) / 2)) / (__d));	\
 }							\
-- 
1.7.9.7